Streaming Object Detection and Segmentation with Polar Pillars

2. The system of claim 1, wherein computing features for a current sector comprises trailing edge context padding, wherein the features from a preceding sector at time t−1 are used to pad the features computed for the current sector at time t.

3. The system of claim 1, wherein computing features for a current sector comprises bidirectional context padding, wherein features from a preceding sector at time t−1 and aggregated features from a full sweep of a previous time frame are used to pad the features computed for the current sector.

4. The system of claim 1, wherein performing simultaneous 3D object detection and segmentation using the rectangular feature map outputs an image of an environment, wherein pixel of the environment includes a Boolean value that indicates if an object is present or not present in the image.

5. The system of claim 1, wherein performing simultaneous 3D object detection and segmentation using the rectangular feature map comprises outputting a bounding box and an object classification associated with each bounding box.

6. The system of claim 1, wherein simultaneous 3D object detection and segmentation is performed by a neural network trained by simulating a streaming dataset.

7. The system of claim 6, wherein the streaming dataset is simulated by artificially slicing input data points into sectors according to their respective azimuth.

9. The method of claim 8, wherein computing features for a current sector comprises trailing edge context padding, wherein the features from a preceding sector at time t−1 are used to pad the features computed for the current sector at time t.

10. The method of claim 8, wherein computing features for a current sector comprises bidirectional context padding, wherein features from a preceding sector at time t−1 and aggregated features from a full sweep of a previous time frame are used to pad the features computed for the current sector.

11. The method of claim 8, wherein performing simultaneous 3D object detection and segmentation using the rectangular feature map outputs an image of an environment, wherein pixel of the environment includes a Boolean value that indicates if an object is present or not present in the image.

12. The method of claim 8, wherein performing simultaneous 3D object detection and segmentation using the rectangular feature map comprises outputting a bounding box and an object classification associated with each bounding box.

13. The method of claim 8, wherein simultaneous 3D object detection and segmentation is performed by a neural network trained by simulating a streaming dataset.

14. The method of claim 13, wherein the streaming dataset is simulated by artificially slicing input data points into sectors according to their respective azimuth.

16. The at least one non-transitory storage media of claim 15, wherein computing features for a current sector comprises trailing edge context padding, wherein the features from a preceding sector at time t−1 are used to pad the features computed for the current sector at time t.

17. The at least one non-transitory storage media of claim 15, wherein computing features for a current sector comprises bidirectional context padding, wherein features from a preceding sector at time t−1 and aggregated features from a full sweep of a previous time frame are used to pad the features computed for the current sector.

18. The at least one non-transitory storage media of any claim 15, wherein performing simultaneous 3D object detection and segmentation using the rectangular feature map outputs an image of an environment, wherein pixel of the environment includes a Boolean value that indicates if an object is present or not present in the image.

19. The at least one non-transitory storage media any of claim 15, wherein performing simultaneous 3D object detection and segmentation using the rectangular feature map comprises outputting a bounding box and an object classification associated with each bounding box.

20. The at least one non-transitory storage media any of claim 15, wherein simultaneous 3D object detection and segmentation is performed by a neural network trained by simulating a streaming dataset.